New research published in the prestigious journal Nature indicates that large language models (LLMs), such as AI chatbots, can offer a reliable source of medical information for the general public. The study, titled 'Reliability of LLMs as medical assistants for the general public: a randomized preregistered study', suggests these artificial intelligence tools could play a supportive role in healthcare by providing accurate answers to common health queries.
Conducted by a team of international researchers, the study utilised a randomised and preregistered methodology to assess the reliability of LLMs when responding to health-related questions typically posed by the public. The findings, which have been peer-reviewed, show that the information generated by these AI models was largely consistent with established medical knowledge, suggesting a level of accuracy that could be beneficial for individuals seeking initial health guidance or general understanding of conditions.
Despite the promising results, the researchers behind the study, whose institutional affiliations were not specified in the provided summary but are detailed in the full Nature publication, strongly caution against using LLMs as a substitute for professional medical advice. They emphasise that while AI can provide information, it cannot diagnose conditions, offer personalised treatment plans, or account for individual medical histories and nuances that only a qualified healthcare professional can assess. The study positions LLMs as potential 'medical assistants' for the general public, aiding in information dissemination rather than clinical decision-making.
This research contributes significantly to the ongoing discussion about the integration of AI into healthcare. Previous studies have explored various applications of AI, from image analysis in diagnostics to administrative support. However, this particular study focuses on the direct interaction between AI and the public for general health inquiries, providing valuable insights into the practical reliability of LLMs in this context. The findings underscore the potential for AI to democratise access to health information, particularly in areas where access to medical professionals might be limited.
For the UK, these findings could have implications for how public health information is accessed and disseminated. With increasing pressure on NHS services, tools that can reliably answer common health questions could potentially reduce the burden on GPs and other healthcare providers by empowering individuals with accurate initial information. However, this also raises important questions about digital literacy, the potential for misinformation if not properly managed, and the need for clear guidelines on the appropriate use of AI in health contexts.